Anthropic, OpenAI and Google DeepMind Are Quietly Building an AI Standards Body

My take: Three laboratories that compete directly in the AI market sitting down to design a standards body is not a small thing. It signals that at least part of the industry acknowledges it cannot keep being judge and jury of its own safety evaluations.

The structure Demis Hassabis is proposing, modeled on the U.S. FINRA, would target independent technical audits before a model ships. That is exactly the kind of external check that is missing today: when a company publishes a safety report about its own model, nobody can independently verify whether the methodology was rigorous.

For any professional or organization using these tools, a credible standard matters: it defines what "safe to deploy in production" actually means. The question is whether this effort will carry real weight, or remain a statement of intent among competitors who have very different incentives when it comes to slowing down.